Part 1: Lecture (public)
Outlier detection is one of the fundamental tasks in data mining, besides clustering, frequent pattern analysis, and classification. In this lecture, we will learn about outlier detection in relation to the fundamental tasks of data mining as well as in its roots in mathematical and statistical research. We will detail classic methods as well as some more recent methods for outlier detection in the data mining literature. We will explore their application to special domains and to data with special challenges such as high dimensional data. An important question will also be how to evaluate, interpret, and make sense out of outlier detection results.
During the lecture, we might occasionally examine the typical behavior of some representative algorithms on toy data sets. Participants interested in following these experiments on their own laptop computer are encouraged to download the latest release of ELKI ( http://elki.dbs.ifi.lmu.de/ ) (requires java, e.g., OpenJDK 7).
The lecture will be presented in 4 units:
18.5., 10-12
19.5., 10-12
19.5., 14-16
20.5., 10-12
Part 2: Seminar, 12 participants
In the seminar, we will discuss recent literature on outlier detection and possible application sin the participants' research domains. The participants are to prepare talks on papers and to discuss the presented papers and present their exploration on the application of outlier detection methods in their research domain, based on the insights learned in the lecture.
The seminar will comprise 6 units:
22.6., 10-12, 14-16
23.6., 10-12, 14-16
24.6., 10-12, 14-16